The Blog
How High School Students Beat the Data Scientists

How High School Students Beat the Data Scientists


How a Bunch of High School Students Beat a Team of Data Scientists

How a Bunch of High School Students Beat a Team of Data Scientists  
By Arijit Sengupta


Here’s one of the biggest myths about AI:

“Only trained data scientists can create reliable AI; ordinary users should step aside and let the professionals handle it.”

My years of working in AI have convinced me that the conventional wisdom is dead wrong. In my experience, even skilled data scientists can create an AI that’s accurate, but is accurately wrong when it comes to solving real- world business problems.

That’s what led us to develop Aible, an intuitive AutoML tool that makes AI accessible to anyone – no previous training or data science experience required. We created Aible to enable ordinary people to create high-quality custom AI models that maximize real-world business impact.


But is Aible really better than a team of data scientists?

To put Aible to the test, we sponsored a competition at the UC Berkeley AI Summit that pitted a group of experienced data scientists against high school and college students with no prior AI training. The challenge: solve an important real-world healthcare problem in just two hours using the AI tools of their choice. The competitors could develop the AI using their favorite data science toolkit or use Aible’s automated machine learning tool.

The participants were asked to analyze a 56,000-patient dataset to determine patient readmission patterns at hospitals. Eleven data scientists used AI that they developed using standard data science and AutoML tools. The high school and college students with no previous training in AI used Aiblel. Some of the students had one hour of training on Aible, while others ran in straight from the soccer field and used Aible without any training at all.

The problem that the teams were asked to solve is one that regularly faces hospitals – how to predict which patient would be readmitted so that the right decisions can be made to optimize both the health outcomes for patients and the cost for hospitals. The cost of a patient being released incorrectly and then readmitted is very high – it not only costs the hospital money, but also puts the patient’s health at risk. But holding a patient back unnecessarily has a cost too. Finally, there are only so many beds in a hospital, so you can’t realistically hold back every patient at risk of readmission. So, hospitals need an AI that takes into account critical real- world factors such as cost/benefit tradeoffs and resource constraints when it predicts readmissions.

The results of the Real-World AI Challenge were a surprise to a lot of people – particularly the data scientists. The top scorers included two high school students and a history major. Not one data scientist made it into the top ten. The winners finished the challenge in as little as 30 minutes after just one hour of training on Aible.

On average, users using Aible were twice as good as the average data scientist. Then the data scientists said, “This is unfair. Two hours is not enough. Give us two days.” We gave them five days. After five days, only four out of 11 data scientists beat the Aible users.

So why was it so hard for the data scientists? A key reason is that it’s easy – almost too easy – for data scientists to build predictive models with off- the-shelf tools and techniques. When it comes time to consider a real-world situation – such as what happens when a patient is unnecessarily discharged from a hospital – they’re out of their element. Suddenly, they have to think about the underlying principles of AI. They can’t rely on standard AI tools and techniques.

Aible on the other hand asks people simple questions to understand the business realities, so it can create a real-world AI that conforms to those constraints. The high school students focused on correctly answering the questions Aible was asking, based on what they observed. The skill the students had to exhibit was curiosity, something we all have, as opposed to the specialized skills of a data scientist. Aible let them create a custom AI that was superior to anything the data scientists built.

For AI to achieve its true potential, users with no previous training have to be able to create custom AI for their own needs. AI needs to speak business; businesspeople shouldn’t have to speak AI. The best AI lets anyone access its power – even high school students.


To learn more about future CDO events, click here.

Share this:
Growing Those Margins CDO 08
Growing Those Margins

By Neil Morling, Chief Financial Officer at Olswang

Data Challenges in the Hotel Industry CDO 26
Data Challenges in the Hotel Industry

Lee Clark, Head of Data and Analytics, Atlas Hotels

Request more information

You have missed out some details, please try again.

Your Name:
Job Title:
Company Name:
Please answer the above question to prove that you are human.

©2021 Global Business Intelligence | All Rights Reserved

We use cookies to ensure that we give you the best experience on our website. If you continue without changing your settings, we'll assume that you are happy to receive all cookies. You can find out how we use cookies here.